Fault detection of lithium-ion battery packs with a graph-based method

Abstract

A fast fault detection of lithium-ion battery (LiB) packs is critically important for electronic vehicles. In previous literatures, an interleaved voltage measurement topology is commonly used to collect working voltage of each cell in LiB packs. However, previous studies ignore the structure information of voltage sensor layout, leading to a large time delay for LiB fault detection. To tackle the above issue, this work proposes a graph-based autoencoder, which uses graph data of voltage sensors to strengthen the reconstruction ability of the traditional autoencoder, and combines the voltage reconstruction errors with a specific detection strategy to identify three common fault types and their locations in LiB packs. Our results obtain the mean AUC of more than 88% for all three fault types and the mean relative detection time of 1.5 s, 11 s and 0.1 s for three fault types, respectively, which perform better than state-of-the-art methods. Moreover, identification of mixed faults is also performed to validate the proposed method, and the results reveal that the proposed method is reliable and fast for fault detection of LiB packs.